the fall of the labor share and the rise of superstar firms€¦ · the fall of the labor share and...

The Fall of the Labor Share andthe Rise of Superstar Firms

David Autor, MIT and NBERDavid Dorn, University of Zurich and CEPRLawrence Katz, Harvard and NBERChristina Patterson, MITJohn Van Reenen, MIT and NBER

National Bank of Belgium, October 15th 2018

Falling Labor Share of Corporate sector Value-Added Evident in Many Countries

Karabarbounis and Neiman, 2014

The Rise of Superstar Firms

Source: Compustat USA

Global Sales of Top500 US Firms tripled from$4tr in 1972 to $12tr in 2015

Top3 in 1985

Top3 in 2015

The Rise of Superstar Firms doesn’t just reflectUS GDP growth


US Sales of Top500 US Firms / US GDP

Significance of decline in Labor share

• Why should we care about fall in Labor Share?1. Overturns a key ‘Kaldor fact’

2. Slow GDP growth → Labor getting a shrinking slice ofslow-growing pie

3. Distribution of capital far more unequal thandistribution of labor → Growing income inequality (IMF,’17)

• Fall is real and significant• Elsby et al. ’13; Karabarbounis & Neiman ’14, ‘18;

Rognlie ’15; Koh et al. ’17; Piketty ’14; Bridgman ’14;Smith et al ’17; Autor & Salomons, ’18

Causes of the Falling Labor Share?

Role of technical change: Karabarbonis & Neiman ‘14• Falling capital price and, critically, elas. of L-K sub > 1

• But empirical literature suggests < 1, e.g., Lawrence ’15,Oberfield-Raval ’14, Antras ’04, Hamermesh ’90

Role of trade exposure: Elsby et al. ’13• Driven by falling labor share in trade-impacted manufacturing

industries (China competition)

These representative firm models underplay fact thataggregate fall reflects reallocation between firms

• Role of rising profit share – higher aggregate mark-up(Eeckhout and de Loecker ’17)

Contributions of this PaperOffers a new ‘Superstar Firms’ hypothesis

• Large firms tend to have lower labor shares• Rising prevalence of “winner take most” competition• Small set of large firms capture increasing share of

market, aggregate labor share falls due to reallocationPresents evidence consistent with this hypothesis

1. Three decades of outcome measures2. U.S. firm & establishment data – Economic Censuses

from multiple sectors (not just manufacturing)3. Cross-national OECD comparisons using industry

(KLEMS, COMPNET) & firm-level (BVD ORBIS) data

Summary of Evidence

1. A rise in sales concentration within four-digitindustries across US private sector

2. Industries with larger increases in concentration seelarger falls in labor share

3. Labor share fall largely due to reallocation ofactivity between firms, not primarily a general fallwithin all firms

4. Reallocation component of falling labor sharelargest in industries with rising sales concentration

5. These patterns are seen internationally, not just inUS

Some Related Literature• General Trends: Piketty ’14; IMF ’17• Explanations of labor share fall: (a) Measurement: Rognlie ’15;

Smith et al ’17; (b) Market Power: Kalecki ‘38; Barkai ’16; Gutierrez& Philippon ’16; Grullon et al ’17; Berkowitz et al ‘17; Eeckhout &De Loecker ’17; Hall ‘18 (c) ICT: Karabarbounis & Neiman ’14, ‘18;(d) Trade: Elsby et al ’13; (e) Regulations & Institutions: Blanchard &Giavazzi ’03; Azmat et al ’12

• “Superstar” Firms: Brynjolfsson & McAfee ’08; Furman & Orszag’15; Bain ‘51; Demsetz ‘73; Schmalensee ’87

• Productivity: Bartelsman, Haltiwanger & Scarpetta ’13; Decker, etal. ’17; Andrews et al ’15;

• Firm heterogeneity & Wage Inequality increase: Davis &Haltiwanger, ’92; Faggio et al, ’10; Card et al ‘13; Song et al ’17

• Firm-level Decompositions of labor share: Bockerman &Maliranta ’12; Kehrig & Vincent ’17; Lashkari & Bauer ‘18

Overview

1. A Model of Superstar Firms

2. Data and Measurement

3. Evidence

4. Discussion

Superstar Firm Model in new Appendix A(Generalization of Melitz & Ottaviano ’08 )

• Monopolistic Competition with heterogeneousfirms─ General class of utility functions consistent

with “Marshall’s 2nd Law of Demand”(generates variable mark-ups unlike CESDixit-Stiglitz preferences)

─ General class of underlying firm productivitydistributions (nests Pareto pdf)

Superstar Firm Model Sketch

Heterogeneous firms in an industry, (TFPQ)• =

‒ = value-added‒ = capital‒ = labor

• Imperfectly competitive product markets with a mark-up ofprice over marginal cost• = /

• Competitive factor markets: wage , capital cost• Firms take random draw of productivity from a distribution

with pdf λ(z). Productivity draw determines firm’sidiosyncratic marginal cost

The Firm-level Labor Share,

Taking FOC with respect to labor gives labor share,= payroll ( ) over value added ( ) for firm

• = =

• More productive/lower marginal cost (high“superstars”) firms have:

‒ larger market share ( =∑

) - more output dueto lower marginal costs

‒ lower labor share ( ) because their mark-up ishigher (e.g. Melitz & Ottaviano ’08; oligopoly modelslike Cournot, etc.). Why?...

Higher mark-up ( ) for more productive firmsarises in many standard cases

1. Demand more inelastic when price is lower. Highly productivefirms charge lower prices & so face more inelastic demand. Thusmark-ups higher

2. In our data we confirm that larger firms have lower labor shares(& higher mark-ups as in de Loecker & Warzynski ’12)

3. Consistent with Pass-through literature: 1% marginal costincrease causes less than a 1% increase in price (e.g. Arkolakis et al,’18 survey)

4. Note: CES preferences imply common mark-up. But if allowfixed costs of labor (Bartelsman et al ’13) = V + , thensuperstar firms still have lower labor shares since

= = +

Change in economic environment• Change in environment which reallocates more market shareto superstar firms will tend to (i) increase concentration and (ii)reduce aggregate labor share. Examples:• Increased importance of platform competition (network

effects, especially in digital markets)

• Larger firms better at exploiting intangible capital; e.g. ICT –Besson ’17; Lashkari & Bauer ’18; Eberly & Crouzet ‘18

• The “Matthew effect” of globalization: allocates more outputto more efficient firms - Melitz, ’03; Mrázová & Neary ’17

• Falling competition? Grullon et al. ’16; Gutierrez & Philippon’17, Döttling et al ‘18 on weaker anti-trust, greater regulation& occupational licensing. But…

Consider increase in market toughness(globalization or higher variety substitutability)

Modelled as a fall in minimum cost threshold to produce inmarket (c*: if a firm’s marginal cost, c > c* it will exit)

1. Output shifts to low labor share firms. “Between firm”reallocation pushes down aggregate labor share

2. But for an individual firm, labor share rises becausemark-up falls (“within firm”)

Increase in market toughness depends on pdf ofproductivity, λ(z)

• Reduces industry labor share if λ(z) is log convex;

• Unchanged if log linear (e.g. Pareto case); Increases if logconcave

Hence, fundamentally an empirical issue

Predictions: Consider a Change in Environmentthat Favors Most Productive/Superstar Firms

1. Concentration levels will increase

2. Industries with largest increases in concentrationwill have biggest falls in labor share

3. Fall in labor share mainly due to reallocationtowards low labor share firms (rather than uniformfall)

4. Rising industry concentration will predict thereallocation component of rising labor share

5. If the underlying forces are global, these regularitieswill be seen in many countries

Overview



3. Evidence

4. Discussion

Data Sources (USA)

Labor share and sales concentration• US quinquennial Economic Censuses, 1982 – 2012• Use six sectors covering ∼ 80% of private sector jobs

1. Manufacturing2. Retail3. Wholesale4. Services5. Utilities & Transportation6. Finance

• 5.2 million establishment-year observations• 4.0 million firm-year observations• Consistent series of four digit SIC codes

Overview



3. Evidence

4. Discussion

Summary of Evidence


2. Industries with larger increases in concentrationsee larger falls in labor share

3. Fall largely due to reallocation of employmentbetween firms not a general fall withinincumbent firms

4. Reallocation component of falling labor sharelargest in industries w/rising salesconcentration

5. These patterns broadly international in scope

Fig 4: Rising Concentration:Manufacturing and Retail Trade

Manufacturing Sector Retail Trade

Notes: Weighted average of 4 digit industries within each large sector. Manufacturing:388 inds; Retail: 58;

CR20

CR4

Fig 4: Rising Concentration:Finance and Wholesale Trade

Finance Wholesale Trade

Notes: Weighted average of 4 digit industries within each large sector. Wholesale: 56;Finance: 31.

Fig 4: Rising Concentration:Services and Utilities + Transport

Service Sector Utilities + TransportationSector

Notes: Weighted average of 4 digit industries within each large sector. Services: 95;Utilities & Transport: 48.

Summary of Evidence






Fig 5: Basic Descriptive Relationship-Larger Firms Have Lower Labor Shares

-2.37

-1.58

-0.94-0.70 -0.64

0.00

-3.50

-3.00

-2.50

-2.00

-1.50

-1.00

-0.50

0.00

0.50

Wholesale

Finance

ManufacturingRetail

Uls+Transport

Services

= + Sales +

Table 2: Rising Concentration and Falling LaborShare; Manufacturing, 5 year changes

∆Payroll

Value Added= ∆ = + ∆Conc + +

Notes: ** significant at 1% level; * = significant at 5% level; ~ = significant to 10% level

Fig 6: ∆Labor Share of Sales regressed on∆Concentration: Results Across Six Sectors

∆ = + ∆ 20 + +

Notes: OLS Regression coefficient of ∆Labor Share (payroll over sales) on CR20 (5year changes); 95% confidence intervals; 1982-2012.

Fig 7: Negative relationship between industrylabor share & CONC gets stronger over time

Notes: OLS Regression coefficient of ∆lab share (payroll over value added) on CR20(5 year changes); 95% confidence intervals; 1982-2012.

Summary of Evidence



3. Labor share fall largely due to reallocation ofactivity between firms, not primarily a generalfall within incumbent firms



Olley-Pakes (1996) DecompositionApplied to Labor Share= ̅ + Σ − − ̅

• S = ∑ is aggregate labor share

• ω =∑

is value added share of firm

• ω & ̅ are unweighted mean• Aggregate labor share divided into:

1. Cross-firm unweighted average, ̅

2. Reallocation (covariance) term Σ − − ̅

• Intuition is that overall labor share depends onwithin firm (unweighted) mean + between firmcovariance (bigger firms have lower labor shares)

Dynamic OP Decomposition between periods 2& 1: Melitz-Polanec ‘15 add Entry + Exit

∆ = − = ∆ ̅ + ∆ Σ − − ̅+ , , − , + , , − ,

1. ∆ ̅ is the change in unweighted mean laborshare within surviving firms

2. ∆ Σ − − ̅ is reallocation betweensurvivors

3. , , − , is contribution of exiting firms4. , , − , is contribution of entering firms

• Also do alternative shift-share decompositions

Fig 9: MP Decomposition for Manufacturing:Between firm reallocation main component

Notes: Overall labor share falls 16.5 percentage points 1982-2012. MP decompositionover 5 year periods, aggregated to two 15 year periods

For Wage Bill over Value Added asLabor Share Measure

Reallocationbetween survivors


Notes: MP decomposition over 5 year periods, aggregated to two 15 year periods



Within firm


Notes: MP decomposition over 5 year periods, aggregated to two 15 year periods



Within firm

Reallocationvia Exit

Reallocation via Entry

∆ Labor-Share Decomposition in 6 Sectors:Reallocation component dominates

Notes: MP decomposition over 5 year periods, aggregated over the full sample period

-2.4%

-3.6%

-4.4%

-0.4%

-5.0%

-4.0%

0.6%

6.3%

4.0%

2.4%

-1.2%

3.7%

-10% -8% -6% -4% -2% 0% 2% 4% 6% 8% 10%

Utils+Transport ('92-'07)

Finance ('92-'12)

Wholesale ('82-'12)

Services ('82-'12)

Manufacturing ('82-'12)

Retail ('82-'12)

Between-Firm Within-Firm Firm Exit Firm EntryUsesPayroll/Sales

∆ Labor-Share Decomposition in 6 SectorsUnweighted mean lab share for incumbents rises

Notes: MP decomposition over 5 year periods, aggregated over the full sample period

-2.4%

-3.6%

-4.4%

-0.4%

-5.0%

-4.0%

0.6%

6.3%

4.0%

2.4%

-1.2%

3.7%

-10% -8% -6% -4% -2% 0% 2% 4% 6% 8% 10%

Utils+Transport ('92-'07)

Finance ('92-'12)

Wholesale ('82-'12)

Services ('82-'12)

Manufacturing ('82-'12)

Retail ('82-'12)

Between-Firm Within-Firm Firm Exit Firm Entry

Lab share generally risingwithin firms

Price-cost markups (Waiting Census clearance)

1. Harder to estimate mark-ups than labor shares!• Sales/Costs (Antras, Fort & Tintelnot ’17)

• Using FOC (de Loecker & Warzynski ’12; Hall ’88)

─ Estimate production function in each industry toobtain elasticity of output wrt to variable factor(α ); divide by factor share ( ):

=

2. Using all methods we observe (in CfM):• Higher mark-ups for larger firms in cross section

• Increase in aggregate mark-up but like labor sharerelatively small change in median & unweighted averagemark-up (again, it’s reallocation)

Summary of Evidence






Fig 11: Regression of ∆Labor Share Components onSector Level ∆ CR20: Loads on reallocation term

-0.40 -0.35 -0.30 -0.25 -0.20 -0.15 -0.10 -0.05 0.00 0.05 0.10

Finance

Services

Utils+Transport

Manufacturing

Retail

Wholesale

Between-Firm Within-Firm Firm Entry Firm Exits

Summary of Evidence


2. Industries with larger increases in concentration seelarger falls in labor share

3. Fall largely due to reallocation of employmentbetween firms not a general fall within incumbentfirms

4. Reallocation component of falling labor sharelargest in industries w/rising sales concentration

5. These patterns are broadly international in scope

Data Sources (International)

Industry-level labor shares, intermediate services• KLEMS data• 11 countries, 32 industries

Industry-level labor shares and concentration• ECB COMPNET data• 14 countries, 53 industries

Firm-level labor shares• BvD Orbis data• 6 EU countries

Fig 12A: Correlations of industry-level Labor ShareLevels Across Countries

Average correlation coefficient from pairwise correlations between indicatedcountry and each of the 11 other countries

Fig 12B: Correlation of Industry Labor ShareChanges Across Countries

Average correlation coefficient from pairwise correlations between indicatedcountry and each of the 11 other countries; fraction of negative correlations

Concentration trends (OECDMultiProd)

Concentration trends (Orbis)

Bajgar, Criscuolo and Timmis (forthcoming): M&As, productivity and concentration, OECD.

Fig 13: ∆Labor Share: Within/Between-FirmDecomposition by Country Using BVD Orbis Data

0.2

-0.4

0.6

-3.9

-0.5

7.6

-1.2

-1.3

-2.7

-5.5

-7.1

-10.4

-12 -10 -8 -6 -4 -2 0 2 4 6 8 10

Portugal ('05-'10)

France ('03-'08)

Italy ('05-'10)

Germany ('05-'10)

UK ('03-'08)

Sweden ('03-'08)

Between-Firm Within-Firm

Table 6: Industry Regs of ∆ Labor Share of Sales on∆ Concentration (COMPNET, 10 year change)

0.33

0.01

-0.04

-0.05

-0.08

-0.10

-0.13

-0.14

-0.15

-0.18

-0.18

-0.20

-0.28

-0.34

-0.60 -0.50 -0.40 -0.30 -0.20 -0.10 0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70

Belgium

Poland

Latvia

Lithuania

Portugal

Slovenia

Estonia

Romania

Germany

Finland

France

Italy

Austria

Slovakia

Overview



3. Evidence

4. Discussion

Summary of Empirical Findings

1. A pervasive fall in labor share across countries

2. Mainly due to reallocation of sales between-firmswithin industries rather than within-firm changes

3. Industries with largest increases in concentrationhad largest falls in labor share

4. And this was due to the reallocation componentof falling labor share, not a general fall in share

5. Comparable international findings in industry &firm-level data across OECD countries

What’s Not Going on

Results do not appear explained by

1. Country-specific institutional factors like specificregulations or weakening labor unions

2. Susceptibility to ‘routine-replacing technicalchange’ (ICT)

3. ‘China shock’ – trade exposure not major predictor

Does Not Appear to be the ‘China Shock’:But Does Not Lower Labor Share

Sample 1: 1992-2012-3.69 ** -0.86 * -1.16 ** 6.70 * 2.46(1.42) (0.36) (0.42) (3.24) (1.83)

-2.67 ** -1.13 ** -1.24 ** 0.32 -1.29(1.00) (0.41) (0.42) (3.24) (1.48)

Dln SalesDln Wage

BillDln Value

AddedD LaborShare

D Payroll-to-Sales

Industry-Level Regressions for Manufacturing: Effect of Change in Chinese ImportExposure on Sales, Wages, Concentration, and Labor Share

Chinese ImportExposure: 5 year D 's

(7) (8)

Chinese ImportExposure: 5 year D 's

Sample 2: 1992-2007

(1) (2) (3)2SLS Estimates

Notes: ~ p ≤ 0.10, * p ≤ 0.05, ** p ≤ 0.01. Regressions reflect 2SLS estimates, using the growth in imports fromChina to 8 other developed countries as an instrument for the growth in Chinese imports to the U.S. (as in Autor etal. 2013) and various industry-level outcome measures, denoted by the column header. Regressions include yeardummies and standard errors are clustered at the slightly aggregated SIC codes, consistent with Autor, Dorn andHanson (2013).

Fig 14: Not Simply “Rigged Economy:” ConcentratingIndustries Show Larger Increase in Innovation, Productivity

Robustness/Extensions1. Outsourcing/Offshoring

• Compustat evidence

2. Productivity

3. Relabeling labor income into capital

4. Compustat analysis

Conclusion: Much Supporting Evidence for‘Superstar Firms’

1. Tougher competition?• More consumer sensitivity to price/quality

2. Shift towards ‘winner take most’ markets?• IP and information-intensive goods

3. Less creative destruction?• Less entry/exit/startup, Decker et al ‘14, Şahin et al ‘17• More persistent tech. leaders, Acemoglu-Hildebrand ‘17• Laggard firms catching up less quickly, Andrews et al, ‘16

4. Does ↑ concentration indicate weaker competition?• Good news: concentrating industries look dynamic• But once dominant, firms can raise barriers to

growth/entry

Back Up

Correcting Census decompositions forintermediate inputs using NIPA

-14 -12 -10 -8 -6 -4 -2 0 2 4 6

Manufacturing

Wholesale

Retail

Services

Utilities+Transport

Finance

Entry Exit Between Within

Notes: MP decompositions over the full sample period. Use NIPA to adjust Census forintermediates

Concentration trends (MultiProd)

Source: IMF (2017) “Gaining Momentum” http://www.imf.org/en/Publications/WEO/Issues/2017/04/04/world-economic-outlook-april-2017#Summary

Industry Codes• “Retail & wholesale” has Office equipment,computers & software nec. 5044/4045/5046

• Services – computer programming & related(7371/7372/7273/7279/7378/7377)

Productivity paradox• If labor share fall was due to a general drop in competitionthis would help explain productivity slowdown.

• However we find unweighted average firm LS/mark-upshaven’t changed much - Reallocation matters more

• But reallocation to more productive firms should generatehigher productivity growth, but growth has actually slowed

• We do see faster productivity growth and innovation in theconcentrating sectors where LS declining

• So culprit for productivity slowdown need to be foundelsewhere than falling competition (finance; uncertainty; ideasharder to find; mismeasurement, etc.)

Decompositions• Labor share decomposition similar to (inverse) laborproductivity decomposition

─ But different from standard TFP decomposition

─ And standard model would have increased labproductivity growth but unchanged lab share (notsecular decline in lab share)

• We find larger role for reallocation than usualdecompositions

Outsourcing/Offshoring• Domestic outsourcing can’t be direct cause of aggregate LSfall - workers show up somewhere (would have to be some fallin rents type story)

• If offshoring was the cause, can assess this by looking atCompustat data – we see fall in even in multinationals (Butcould be offshoring AND outsourcing (e.g. Apple/FoxComm)

• Payroll/sales could fall with outsourcing, but no obvious biaswith payroll/value added (VA net of intermediate inputs)

• Control for underestimated service inputs by looking withinSIC4 for decompositions

• Underestimate Δoutsourcing for large firms? Implies bigwithin firm fall in LS. We don’t see this.

Concentration & Labor Share: Magnitudes• Counterfactual: If concentration had stayed at 1982 levelswhat would the labor share have been in a sector in 2012compared to actual level?

• Example of CR20 (see Figure 7)

• Varies from 10% in manufacturing to 100% in retail

• Surprisingly low in manufacturing? Effect increased over timeas coefficient on concentration rises. In the last 15 years 1997-2012 over 1/3 of change accounted for

The Rise of Superstar Firms


Dispersion of Sales among Top 500 Firms

Change in the Labor Share in US manufacturing

US Labor Share 1947-2016

Source: BLS https://www.bls.gov/opub/mlr/2017/article/estimating-the-us-labor-share.htm

NIPA vs Census. Manufacturing Labor Share

Concentration trends (MultiProd)

the fall of the labor share and the rise of superstar firms€¦ · the fall of the labor share and...

Documents